Climate-Informed flood damage assessment in the cropland area across the midwestern USA

In this study, we developed a climate-informed convolutional neural network (CNN) model to estimate flood damages (in acres) in cropland areas (corn and soybean) across the midwestern USA. We first trained and evaluated the CNN model using gridMET datasets from 2008 to 2020, which serve as the reference dataset for downscaling the Coupled Model Intercomparison Project 5 (CMIP5) projections. We then applied the downscaled climate variables in the CNN model to estimate crop damages for the historical baseline period (1976-2005) and the future mid-century period (2041-2070). Results indicate a wide range of damages, spanning from a - 40% to a + 120% by mid-century in the Midwestern counties of the United States. Most climate models project higher damages in Iowa during the early season and lower damages in Minnesota counties during late-season flooding. Due to the varying trends of the environmental variables in the climate models, our model shows discrepancies in projected crop damages. Despite these uncertainties, the findings of this study provide valuable insights into potential future patterns of flood-related crop damages.